<<

Volume 23 (July 2013), 161–165

Herpetological Journal FULL PAPER

Published by the British Density of an environmental weed predicts the Herpetological Society occurrence of the king brown ( australis) in central

Peter J. McDonald1, 2 & Gary W. Luck3

1Flora and Fauna Division, Department of Land Resource Management, Government, , NT, Australia,

2School of Environmental Sciences, Charles Sturt University, Albury, NSW, Australia,

3Institude for Land, Water and Society, Charles Sturt University, Albury, NSW, Australia

The (Pseudechis australis) is a large and highly venomous elapid, which occurs throughout much of mainland Australia. Although an ecological generalist, anecdotal evidence suggests that individuals in the arid-zone are more frequently observed in proximity to dense grass cover. We tested the hypothesis that P. australis are more likely to be located close to dense grass cover in an arid region near Alice Springs in the Northern Territory. We focused on the environmental weed buffel grass ( ciliaris L.) because this comprises the highest density of cover in the region. Under an Information- Theoretic framework we used logistic regression to model the occurrence of P. australis against a range of habitat variables expected to influence the distribution and abundance. There was substantial support for our hypothesis with the model including only the variable buffel grass as the best ranked model predicting P. australis presence. The probability of recording P. australis in a location increased with the density of buffel grass cover. Eradicating or reducing buffel grass in and around built-up areas may reduce the risk of interactions between humans or domestic and P. australis.

Key words: elapid, habitat, snake, venomous

INTRODUCTION The highly readily out-competes native vegetation, frequently forming dense swathes where he king brown or mulga snake (Pseudechis australis) is native vegetation had previously been sparse or only Ta large (SVL up to 2.3 metres), heavily-built and highly seasonally dense (e.g., rainfall-triggered growth of native venomous elapid () distributed throughout tussock grasses and forbs) (Clarke et al., 2005; Miller et much of mainland Australia (Shine, 1987; Wilson & Swan, al., 2010; Fig. 1). Although buffel grass tends to favour 2010; Fig. 1). It is regarded as an ecological generalist; and more readily colonize ‘run-on’ landforms (e.g., occurring in habitats ranging from tropical woodlands in floodplains), it is increasingly adapting to and spreading the northern parts of its range to arid spinifex deserts in less productive areas (Smyth et al., 2009; Miller et al., in the south and feeding on a variety of terrestrial 2010). vertebrate prey (Shine, 1987; Wilson & Swan, 2010). Given the continuing spread of buffel grass throughout Although considered as a habitat generalist, anecdotal arid Australia, the propensity for it to form dense stands, evidence suggests that arid-dwelling P. australis are more and the possible association between P. australis frequently observed in proximity to dense , occurrence and grass density, we suggest that buffel grass particularly when these grasslands occur in riparian may be an important environmental factor influencing habitats (Orange, 2009; PM pers. obs.). the distribution and abundance of the snake species. Buffel grass ( - nomenclature Our hypothesis is that the probability of occurrence according to Albrecht et al., 1997) is an exotic perennial of P. australis will increase with increasing buffel grass tussock grass (originally from and south-west density. Such an association has substantial implications ) deliberately established throughout Australia’s for the safety of humans and domestic animals in arid arid zone as pasture for cattle and as a soil stabilizer regions, given that P. australis is a highly venomous and (Smyth et al., 2009). Since its introduction in the late dangerous species (Currie, 2004). Hence, the way buffel 1800s, the grass has spread throughout the northern grass is managed may have major implications for the and central regions of arid Australia and is now regarded outcomes of human-snake and domestic -snake as a serious environmental weed (Eyre et al., 2009). interactions Correspondence: Peter McDonald ([email protected])

161 P.J. McDonald & G.W. Luck MATERIALS AND METHODS set of models considered) of variation in the response variable. Using information from the literature and our Our study was conducted in the MacDonnell Ranges own field observations, we developed a set of models bioregion which covers 39,300 km2 of upland in the to explain the occurrence of P. australis that included southern Northern Territory (NT), Australia (see variables with the potential to influence the distribution McDonald et al., 2012 for a description of the landforms or abundance of the species (see Table 1 for justification and land-uses of the area). We used systematic road- for variable inclusion). Relevant variables were recorded cruising (Rosen & Lowe, 1994) to sample P. australis within a 50 m radius around each location where a along a 77 km sealed road transect which included snake was encountered and a single proximity variable part of Namatjira Drive (23°48’33”S, 133°13’16”E to (distance to major drainage) was also recorded (Table 1). 23°40’1”S, 132°38’8”E) and all of the Ormiston Gorge All variables were also recorded at 50 randomly selected access road (23°41’5”S, 132°42’34”E to 23°37’57”S, locations where the snake species was absent along the 132°43’39”E) west of the town Alice Springs (Fig. 1). This road transect. The randomly selected locations were a transect runs through a range of habitat types typical minimum of 500 m from the nearest P. australis presence of the bioregion (see McDonald et al. 2012 for a map location and can be referred to as ‘pseudo absences’ illustrating the typical mosaic of vegetation communities (Milne et al., 2005). that occurs along the road transect), including areas with We initially tested for collinearity among the varying levels of buffel grass infestation. One of us (PM) independent (predictor) variables using the Spearman drove this transect on 77 nights over a 12 month period correlation coefficient. Because no pair of variables between August 2009 and July 2010. Each night the exhibited a correlation coefficient of greater than 0.6, no transect was driven twice (two laps) at a speed of 40–60 variable was removed from further analysis (McDonald km/hr, with the transect start point alternated between et al., 2012). We further checked for multi-collinearity the east and west ends. The location of all P. australis among the independent variables by regressing each individuals encountered was marked with a hand held independent variable against all the others using linear GPS. All live animals were caught, individually marked regression and examining the values of the variance by scale clipping (Brown & Parker, 1976) and released on inflation factors. In these analyses, all variance inflation the road verge adjacent to the point of capture. Road-kill factor values were less than 2 indicating very low levels animals were removed from the road surface. of multi collinearity (see Zuur et al., 2010). In order to test whether there was an association All variables were modelled using binary logistic between buffel grass density and P. australis occurrence regression, with presence (1) or absence (0) as the in our study area, we modelled data under an dependent variable. We also modelled combinations of Information-Theoretic framework (Burnham et al., 2011). independent variables where these combinations made This involved developing a set of a priori hypotheses biological sense (e.g., RIPAR_50 + MAJDRAIN_DIST). (models) to explain variation in the response variable Models were ranked using Akaike’s Information Criterion

(snake presence/absence) based on consideration of corrected for small sample size (AICc). This is appropriate existing knowledge from the literature and/or field when the number of data points/maximum number of experience (Burnham et al., 2011). Models were then fitted parameters is less than 40 (Symonds & Moussalli,

compared to identify the best explanation (among the 2011). Models with smaller values of AICc have greater Table 1. Habitat variables recorded at presence and pseudo-absence locations for Pseudechis australis

Variable Description Method of data Biological knowledge of P. australis collection MAJDRAIN_DIST Distance (m)to nearest major Google Earth1 Encountered more frequently in close proximity drainage line, identified by presence to alluvial systems (floodplains, rivers, creeks)a, b of River Red gums (Eucalyptus camaldulensis) RIPAR_50 % area of riparian woodland/ Field survey2 Encountered more frequently in close proximity within a 50 m radius of to alluvial systems (floodplains, rivers, creeks)a, b sample site BUFF_COVER % area of Cenchrus ciliaris cover Field survey2 Encountered more frequently in close proximity within a 50 m radius of sample site to dense grasslandb TUSSCK_COVER % area of native soft grass (e.g. Field survey2 Encountered more frequently in close proximity Aristida pp.) cover within a 50 m to dense grasslandb radius of sample site HUMCK_COVER % area of native hummock grass Field survey2 Encountered more frequently in close proximity (Triodia spp.)cover within a 50 m to dense grasslandb radius of sample site References: aPM pers. obs., 2005–2010; bOrange, 2009; Methods: 1Distance tool used on Google Earth Pro, with major drainage lines clearly visible on SPOT-derived satellite imagery; 2On-ground survey with measurements of habitat made by visual estimation

162 Grass weed predicts snake occurrence

Table 2. AICc ranked models explaining the occurrence of Pseudechis australis in the study area

Hosmer and Lemeshow test

a 2 Model (coefficient; standard error) AICc ∆i wi χ p BUFF_COVER (0.06; 0.02) 100.41 0.00 0.31 2.60 0.75 RIPAR_50 (0.01; 0.01) + BUFF_COVER (0.04; 0.02) 100.81 0.40 0.25 4.00 0.67 RIPAR_50 (0.04; 0.02) 101.57 1.16 0.17 0.57 0.45 MAJDRAIN_DIST (-0.00;<0.00) + BUFF_COVER (0.06; 0.02) 102.58 2.16 0.10 8.20 0.41 MAJDRAIN_DIST (<0.00;<0.00) + RIPAR_50 (0.02; 0.01) 103.59 3.17 0.06 10.70 0.22 RIPAR_50 (0.02; 0.01) + TUSSCK (<0.00; 0.03) 103.73 3.31 0.06 9.10 0.17

Constant only (null) model 105.94 - - - - a Models include the 95% confidence set with models ranked by AICc values. Also shown are the criterion values (∆i), 2 Akaike’s weights (wi) and the χ , p-values from the Hosmer and Lemeshow goodness-of-fit tests and theAIC c value of the constant only model. All numbers are rounded to two decimal places. AICc, Akaike’s Information Criterion; MAJDRAIN_ DIST, distance (m) to nearest major drainage line; RIPAR_50, area of riparian woodland/grassland within a 50 m radius; BUFF_COVER, % area of Cenchrus ciliaris cover within a 50 m radius; TUSSCK_COVER, % area of native soft grass cover. support as explanations for variation in the response present the 95% confidence set of models, where the variable relative to other models in the set considered. summed wi equals a minimum of 0.95. To test the overall We assessed the relative strength of models subsequent goodness of fit of each model, we applied the Hosmer to the best fitting model by comparing the difference in and Lemeshow statistic. Significant p-values (p<0.05) criterion values of the best ranked model (smallest AICc from this test are evidence of lack of fit (Quinn & Keough, value; AIC ) with model i (∆ ) (Symonds & Moussalli, cmin i 2002). AICc values, difference in criterion values (∆i), 2011).The best ranked model has a ∆i value of 0 and Akaike weights and diagnostic measures were calculated subsequent models are scored as ∆i=AICci- AICcmin, where manually. All other analyses were run in SPSS (PASW

AICci is the AICc value of the model being compared with Statistics v.17.0). the best ranked model. Models with ∆i values less than 2 are considered to be essentially as good as the best RESULTS model in explaining variation in the response,∆ i values up to 6 should be considered plausible explanations,i values Over the 12 month period we encountered 29 P. australis between 7 and 10 may be rejected, and ∆i values greater on the road transect. Of these individuals; 27 were live than 10 should be considered implausible and rejected and two were road-kills, 17 were males (mean SVL=1042 (Burnham & Anderson, 2002; Symonds & Moussalli, ±49 mm) and 12 were females (mean SVL=810±34 mm).

2011). We also calculated Akaike weights (wi) for each These individuals do not include animals that were model, which can be interpreted as the probability of recaptured (n=2) and only the original capture locations a model being the best model of those considered. We were included in the habitat analysis.

A C

B D

Fig. 1. A) Location of the study area in the Northern Territory, Australia; B) location of the road transect; C) the king brown snake (Pseudechis australis); D) typical buffel grass (Cenchrus ciliaris) infestation along the road transect, note the dense groundcover.

163 P.J. McDonald & G.W. Luck

1 habitats (the lowest lying and therefore most productive 0.9 habitat types in all catchments of the study area) would 0.8 be more strongly related to P. australis occurrence 0.7 than buffel grass itself. Although there was a positive 0.6 0.5 correlation between buffel grass cover and area of 0.4 riparian habitat within a 50 m radius of sample sites 0.3 (rs=0.51, p≤0.001, n=79), there were still substantial areas Probability of occurrence of Probability 0.2 of dense buffel grass in the three other broad habitat 0.1 0 types of the study area (chenopod shrublands, acacia 0 5 10 15 20 25 30 35 40 45 shrublands and hummock grasslands; see McDonald et % cover buffel grass al. 2012), and P. australis was regularly encountered in Fig. 2. Relationship between predicted probability of P. these habitats. This suggests that buffel grass cover itself, australis occurrence and % cover of buffel grass within rather than productivity, may be the more important a 50 m radius. Dashed lines represent 95% confidence factor influencing P. australis occurrence. intervals. The establishment and domination of buffel grass invariably results in increased groundcover and this is

The best AICc ranked model included only the variable particularly pronounced in areas that were previously BUFF_COVER. Pseudechis australis were more likely to dominated by relatively short-lived native tussock grasses be located in areas with increased buffel grass cover and forbs (Clarke et al., 2005; Miller et al., 2010). As a (Table 2; Fig. 2). This variable was also present in the perennial, the above-ground tussock structure of buffel second and fourth best ranked models. With a total wi grass tends to persist longer into low rainfall periods of 0.66 across these three models (1st, 2nd, 4th), together than the annual grasses and forbs (Clarke et al., 2005). with no evidence of a lack of fit, there is substantial In addition, the post-fire response of buffel grass is faster support for BUFF_COVER being a good explanation for than most native grasses (particularly compared with variation in the occurrence ofP. australis (Table 2). There hummock grasses) and, provided soil moisture is present, was also some support for RIPAR_50 as an explanation can quickly regenerate and return to pre-fire densities of P. australis occurrence. This variable was present in (PM, pers. obs.). Together, these factors may result in a four of the six models within the 95% confidence set, preference for buffel grass among those species that are with a combined wi of 0.54 (Table 2). The other alluvial advantaged by dense grass cover, including P. australis. variable (MAJDRAIN_DIST) was present in two models If P. australis preferentially select areas of dense in the 95% confidence set with a combined wi of 0.16. buffel grass, then this may be another example of Pseudechis australis were more likely to be located in a snake species being ecologically advantaged by closer proximity to major drainage lines, although this an anthropogenic land-cover change. For example, relationship was relatively weak (Table 2). Löwenborg et al. (2010) demonstrated that the grass Of the two remaining grass cover variables, HUMCK_ snake (Natrix natrix) has benefited from the presence COVER was absent from the 95% confidence set and of manure heaps in rural areas in . Decomposing TUSSCK_COVER was only present in the lowest ranked manure heaps on farms provide the snakes with superior model with a wi of 0.16. oviposition sites. Higher temperatures, as found within manure heaps, typically resulted in shorter incubation DISCUSSION periods, higher hatch success rates, and larger and faster offspring (Löwenborg et al., 2010). This has effectively The results of the modelling support our hypothesis that enabled the grass snakes to penetrate into regions too P. australis exhibit a preference for areas with dense cold for other oviparous species. Although we are buffel grass. Although there were substantial areas of not suggesting that buffel grass has resulted in an overall both spinifex hummock grassland and native tussock expansion in the distribution of P. australis, it is possible grassland along the road transect (McDonald et al., that the presence of this grass has led to an increase in 2012), there was little or no evidence to indicate that the abundance of the snake in locations where it was either of these grass types influence the occurrence of formerly uncommon or absent. P. australis. This suggests one of two things: i) that P. Although P. australis appears to have benefited australis prefer dense buffel grass for what it provides from the expansion of buffel grass, other weedy grass (e.g., cover); or ii) that the association with dense buffel species have been shown to have a negative impact. For grass is an artefact of additional factors not recorded in example, in Wisconsin (USA), Butler’s gartersnake was this study. found less frequently in areas dominated by the invasive In considering the second explanation, because buffel wetland grass, Phalaris arundinacea, probably because grass tends to occur in areas with run-on hydrology the grass has reduced the area of habitat suitable for (Miller et al., 2010), it is possible that these areas are this rare snake species, although it was unclear why the more productive than the surrounding landscape and snake was avoiding Phalaris infestations (Kapfer et al., that it is this productivity per se that attracts P. australis. 2013). In Utah (USA), the occurrence of two sympatric However, if the association between P. australis and snakes was negatively associated with the density of the buffel grass was related simply to productivity, then it weed cheatgrass (Bromus tectorum), possibly because should follow that the riparian woodland and grassland the grass impeded the mobility of the snakes or dense

164 Grass weed predicts snake occurrence infestations supported fewer prey species (Hall et al., study in the “” of the Northern Territory. Medical 2009). Journal of Australia 181, 693–698. Investigation of the factors driving the association Eyre, T.J., Wang, J., Venz, M.F., Chilcott, C. & Whish, G. (2009). between P. australis and dense buffel grass is beyond the Buffel grass in ’s semi-arid woodlands: response scope of our study and requires further testing (possible to local and landscape scale variables, and relationship with hypotheses include increased protection from predators grass, forb, and reptile richness.The Rangeland Journal 31, or increased food resources). We also recognize that 293–305. intraspecific habitat selection in snakes can vary Hall, L.K., Mull, J.F. & Cavitt, J.F. (2009). Relationship between geographically, seasonally and ontogenetically (Reinert, cheatgrass coverage and the relative abundance of snakes 1993) and that our single study area and relatively small on Antelope Island, Utah. Western North American sample size precludes analysis of these factors. However, Naturalist 69, 88–95. regardless of what is driving the association, our results Kapfer, J.M., Doehler, K. & Hay, R. (2013). The influence of have important implications for mitigating the threat habitat type and the presence of an invasive wetland to humans and domestic animals posed by P. australis. (Phalaris arundinacea) on capture rates of sympatric rare Buffel grass is well established in and around Alice and common gartersnake species (Thamnophis butleri and Springs and in many remote Aboriginal communities Thamnophis sirtalis). Journal of Herpetology 47, 126–130. across the southern NT and northern South Australia. Löwenborg, K., Shine, R., Kärvemo, S. & Hagman, M. (2010). A recognition that P. australis may be more likely to be Grass snakes exploit anthropogenic heat sources to encountered in association with this grass means that overcome distributional limits imposed by oviparity. efforts to eradicate or reduce the density of buffel grass, Functional Ecology 24, 1095–1102. particularly in proximity to dwellings, could reduce the McDonald, P.J., Luck, G.W., Pavey, C.R. & Wassens, S. (2012). risk of interactions between people or domestic animals Importance of fire in influencing the occurrence of snakes and this venomous and dangerous elapid. in and upland region of arid Australia. Austral Ecology 37, 855–864. ACKNOWLEDGEMENTS Miller, G., Friedel, M., Adam, P. & Chewings, V. (2010). Ecological impacts of buffel grass (Cenchrus ciliaris L.) invasion in This project was partially funded by an operating grant central Australia - does field evidence support a fire- through Charles Sturt University. Dr Chris Pavey and invasion feedback? The Rangeland Journal 32, 353–365. Dr Skye Wassens provided advice on study design and Milne, D.J., Fisher, A., Rainey, I. & Pavey, C.R. (2005). Temporal assisted with logistic support. We thank the people who patterns of bats in the Top End of the Northern Territory, volunteered their time to assist with the snake sampling, Australia. Journal of Mammalogy 86, 909–920. including Kelly Knights, Katherine Williams, Gareth Catt, Orange, P. (2009). A specialist generalist? Notes on the diet Greg Fyfe, Paul Gardner, Simon Rathbone and Peter and behaviour of the mulga snake Pseudechis australis Nunn. All snake handling and processing procedures were (Elapidae) in the Kambalda region of Western Australia. approved by the Charles Sturt University Animal Care Herpetofauna 39, 7–13. and Ethics Committee (approval number 09/064) and the Quinn, G.P. & Keough, M.J. (2002). Experimental Design and Department of Natural Resources, Environment, the Arts Data Analysis for Biologists. Cambridge: Cambridge and Sport/NT Parks and Wildlife Service (research permit University Press. number 35656). Reinert, H.K. (1993). Habitat selection in snakes. In Snakes: REFERENCES Ecology and behaviour, 201–240. Seigel, R.A and Collins J.T. (Eds). New Jersey: The Blackburn Press. Albrecht, D.E., Duguid, A.W., Latz, P.K., Coulson, H. & Barritt, M.J. Rosen, P. C. & Lowe, C. H. (1994). Highway mortality of snakes (1997). Checklist for the Southern Bioregions in the Sonoran Desert of southern Arizona. Biological of the Northern Territory: Nomenclature, Distribution and Conservation 68, 143–148 Conservation Status. Alice Springs: Parks and Wildlife Shine, R. (1987). The evolution of : ecological correlates Commission of the Northern Territory. of reproductive mode within a of Australian snakes Brown, W.S. & Parker, W.S. (1976). A ventral scale clipping system (Pseudechis: Elapidae). Copeia 1987, 551–563. for permanently marking snakes (Reptilia, Serpentes). Smyth, A., Friedel, M. & O’ Malley, C. (2009). The influence of Journal of Herpetology 10, 247–249. buffel grass (Cenchrus ciliaris) on biodiversity in an arid Burnham, K.P. & Anderson, D. (2002). Model Selection and Australian landscape. The Rangeland Journal 31, 307–320. Multimodel Inference: A Practical Information-Theoretic Symonds, M.R.E. & Moussalli, A. (2011). A brief guide to model Approach. 2nd edn. New York: Springer Verlag. selection, multimodel inference and model averaging in Burnham, K.P., Anderson, D.R. & Huyvaert, K.P. (2011). AIC behavioural ecology using Akaike’s information criterion. model selection and multimodel inference in behavioural Behavioural Ecology and Sociobiology 65, 13–21. ecology: some background, observations, and comparisons. Wilson, S. & Swan, G. (2010). A complete guide to of Behavioural Ecology and Sociobiology 65, 23–35. Australia. Sydney: New Holland Publishers. Clarke, P.J., Latz, P.K. & Albrecht, D.E. (2005). Long-term changes Zuur, A.F., Leno, E.N. & Elphick, C.S. (2010). A protocol for data in semi-arid vegetation: Invasion of an exotic perennial exploration to avoid common statistical problems.Methods grass has larger effects than rainfall variability. Journal of in Ecology and Evolution 1, 3–14. Vegetation Science16, 237–248. Currie, B.J. (2004). Snakebite in tropical Australia: a prospective Accepted: 14 May 2013

165